US20110078389A1 - Managing and implementing metadata in central processing unit using register extensions - Google Patents

Abstract

A set of default registers of a processor are expanded into metadata registers on the processor of a computer system. The default registers having stored thereon data, while metadata which is related to the data is stored separately on the metadata registers.

Description

FIELD
• The field relates generally to the computing sciences and, more particularly, to metadata in central processing unit (CPU).
BACKGROUND
• Modern programming languages such as C and C++ do not use metadata. Crucial metadata such as bounds for pointers (e.g., bounds refer to the beginning and ending addresses, i.e., lower and upper bound respectively, of the objects pointed to by a pointer) and data types for objects are not available during the execution of these programming languages, such as C/C++ programs. The lack of such metadata for a running software program often results in incorrect behavior of the program and leads to vulnerabilities. One conventional approach to add metadata is to store it adjacent to the data itself in memory, so that software or hardware can locate it; however, this requires a change in memory layout/Application Binary Interface (ABI). As a result, metadata is only used in a very specialized environment where the complete software is required to be recompiled to adhere to the new modified conventions.
• Alternately, software stores the data and metadata in non-adjacent memory locations and further implements a map function to match the data with the corresponding metadata. In this case, the data is loaded into registers, while the corresponding metadata is loaded into other existing registers. Further, the software is required to keep track of the association between the data and the metadata in existing registers and perform the necessary propagation and invalidation of the metadata to keep it in sync with its corresponding data. In this case, when data is passed to functions in register parameters, the corresponding metadata also needs to be passed in registers. This results in lack of transparency and necessitates an extension to the calling convention used. Also, the implementation of this technique is inherently slow because of a high overhead it places on the software and the system, such as added register pressure, additional instructions that are needed for metadata propagation and invalidation, etc., to the point that the technique is not used in practice.
• FIG. 1 illustrates a simplistic view of aprocessor100 havingdefault registers102. A processor, such as a general purpose processor or a special purpose processor, is a semiconductor chip having electronic circuitry designed to process program code instructions. Most processors typically include a number of common architectural features, such asdefault registers102. A typical register provides a small amount of storage space on the processor so certain data can be accessed using the register as opposed to accessing memory which could be a slow process; for example, a register can be used to store instruction input operands and instruction output results.
• Other common architectural features may include a cache to locally store instructions and/or data, fetch circuitry to fetch instructions from cache and/or memory, write-back circuitry to store instruction results into cache and/or memory. Further, theprocessor100 havingregisters102 is shown to be in communication with Random-Access Memory (RAM)104 and Read-Only Memory (ROM)106.
BRIEF DESCRIPTION OF THE DRAWINGS
• Embodiments of the present invention are illustrated by way of example and not by way of limitation in the figures of the accompanying drawings, in which like references indicate similar elements and in which:
• FIG. 1 illustrates a traditional processor having default registers;
• FIG. 2 illustrates an embodiment of a processor having default registers and bound registers;
• FIG. 3 illustrates an embodiment of default registers coupled with bound registers in a processor;
• FIG. 4 illustrates an embodiment of a process for managing metadata using bound registers as extension registers of default registers having data associated with the metadata;
• FIG. 5 illustrates an embodiment of a process for bound check;
• FIG. 6 illustrates an embodiment of a process for managing metadata using bound registers as extension registers of default registers having data associated with the metadata; and
• FIG. 7 illustrates an embodiment of a computing system.
DETAILED DESCRIPTION
• FIG. 2 illustrates an embodiment of aprocessor200 having default (i.e., existing)registers202 andbound registers204. In one embodiment,processor200 is shown to havedefault registers202 andbound registers204 which represent the hardware register extensions of the default registers. Thesebound registers204 provide the hardware register extension to hold metadata (e.g., bounds in bound registers204) and maintain the association of the metadata with its corresponding data (e.g., pointers in default registers202) when it is copied or modified without the additional pressure on thedefault registers202. Also, by providing (implicit or explicit) association between thedefault registers202 and thebound registers204, the calling convention of theseregisters202 remains the same. The metadata being held at thebound registers204 may be used for different purposes by theprocessor200. For example, the metadata include data type, and theprocessor200 may validate the data type before operating on them.
• A register may include a general purpose or integer register (that is used for storing data and addresses and is regarded as a combined data/address register) and other registers, such as a user-accessible register, a data register, an address register, a condition register, a floating point register, a constant register, a vector register, a special purpose register (including an instruction register), a model-specific register, a control and status register, etc. In one embodiment, integer registers are associated withbound registers204 but the embodiments do not limitbound registers204 to integer registers and that any type or a combination of types of registers may be employed. Theregister extensions204 hold metadata associated with data in theregisters202. In one embodiment, when a (general purpose)register202 contains a memory address (e.g., a pointer), and theregister extension204 may be used to specify the range of memory addresses that can be accessed (load/store) usingregister202. The address inregister202 may be outside the bounds specified in itsbound register204. These bounds may be checked at the time of memory access. Each memory address range consists of a lower bound and an upper bound (further described and discussed later in this document). In this case, theregister extension204 is referred to as bounds register or register descriptor.Processor200 further includes anaddress generator unit206 to generate or determine memory addresses. The address might be generated usingregister202 and optionally other registers (e.g., index register in Intel processor and a constant displacement).Processor200 further includes an address comparator (also referred to as bounds check unit)208 to compare lower and upper bounds of address ranges, and amemory access unit210 to provide access to the memory coupled with theprocessor200 and storage mediums within theregisters202,204. As illustrated, thedefault registers202 are coupled with theaddress generator unit206, while both theaddress generator unit206 and thebound registers204 are coupled with the address comparator (computing logic)208 which is further coupled with thememory access unit210.Processor200 may contain other logics, circuits, and modules to perform various tasks of the embodiments as described throughout this document.Default registers202, depending on the nature of tasks being performed, may contain any number and type of registers, such as general purpose registers, floating point registers, etc. For example, relating to an instruction, source registers and destination registers are employed and these may include general purpose registers or floating point registers, etc. So, an instruction may be provided through a source register, while the results of the instruction may be placed at a destination register; the source and destination registers being part of thedefault registers202.
• In one embodiment, a new program instruction is defined to create bounds and store them in thebound registers204. This program instruction can be an entirely new opcode that was previously undefined in a program, or could also be aliased to using previously unused non-operaton (NOP) instruction. The propagation of metadata to maintain consistency is performed by theprocessor200 either implicitly for a subset of instructions that are typically used with pointer manipulation (such as ADD, SUB, INC, DEC, etc) or explicitly using new instructions that indicate propagation behavior. It describes how theprocessor200 preserves the metadata when the data it corresponds to is modified in certain conditions as well as how theprocessor200 destroys the metadata when the data it corresponds to is modified in other conditions. Because in one embodiment, the semantics of existing instructions are employed to propagate, preserve, and validate/invalidate metadata, there are no additional instructions needed in the software program which can potentially negatively impact the program code size and its performance.
• In one embodiment, bound checks are applied to any number of memory instructions and may be explicitly specified via a prefix of an instruction, such as a segment override prefix. When an instruction has the segment override prefix which is the same as its default segment, the instruction may enforce the bound check while, in another embodiment, a completely new prefix may be added to the instruction. The default segment prefix is NOP in those processors that do not implement this technology. The use of NOP prefix allows the software to work correctly without the benefit of metadata on those processors that do not support/implement metadata. In one embodiment, an explicit bounds check instruction may be allowed, where such an instruction would have two or three operands. For example, in a two operand form, the first operand is used as a memory address and the second operand is used as an access size (byte, word etc.).
• FIG. 3 illustrates an embodiment ofdefault registers202 coupled withbound registers204 in a processor. In one embodiment, register extensions (e.g., bound registers)204 may include any number and type of registers based on the number and type ofdefault registers202 as well as on the workings of the processor architecture of the computer system. In this illustration, the register extensions are 128 bits wide with the lowerbound extensions302 being 64 bits wide to hold lower bounds and the upper boundextensions304 being 64 bits wide to hold upper bounds.
• For example, a new instruction, e.g., CPMKDSC r, m, takes a register and an effective address as operands. The register value is the lower bound and the effective address is the upper bound. The instruction writes the lower bound value to the lowerbound register extension302 and the upper bound value to the upper boundregister extension304 corresponding to the specified register. So, for example, CPMKDSC RAX, [RBX] where RAX=1000 and RBX=1004, the lowerbound extension306 will be written with 1000 as the lower bound value, while the upper boundextension308 will be written with 1004 as the upper bound value in RAX's descriptor. This establishes 1000 and 1004 as the lower and upper bounds for the pointer in RAX. In another embodiment, the instruction can have three operands to calculate lower bound and upper bound, while, in yet another embodiment, there can be two separate instructions to generate the lower and upper bound separately.
• Instructions that may be used to perform arithmetic operations on pointers such as INC, DEC, AND, OR, ADD, and SUB preserve the descriptor contents of the register used in the operation. These operations modify the pointer value but do not destroy it which is why the descriptor contents are still valid. Continuing with the above example, if INC RAX is executed five times then the value of RAX will be 1005 and let's say RAX is now used to access memory using MOV RCX, [RAX]. The metadata associated with the pointer in RAX are the lower and upper bounds namely 1000306 and 1004308. This information when matched against RAX being 1005 concludes that an out-of-bounds memory access, i.e. a buffer overflow, is detected. In this embodiment, the CPU/processor may signal a buffer overflow exception so that the programming software can take appropriate corrective actions.
• Instructions such as MOV or LEA are used to copy pointers between registers and are used to propagate contents of a source descriptor to a destination descriptor in addition to copying pointers. This corresponds to copying the bounds in the boundregisters204 when pointers are copied between registers in the default registers202. Any other instruction that does not require or expect metadata may invalidate or destroy the descriptor contents of any destination registers used in instructions; for example, MOV RAX, 0 destroys the descriptor contents of RAX because bounds may be meaningful for pointers and not for integers.
• In one embodiment, software or hardware or a combination of the two may be used to assert a fine grain access control to memory. Access control, in general, refers to the checks performed when various subjects access objects in making sure that any subjects attempting to access objects have permissions to access such objects. For example, pointers can be regarded as the subjects trying to access different objects in memory. Permissions are embodied as bounds associated with the pointer. Access control information is herein referred to as (lower/upper) bound information or values, while access control check is referred to as bounds check.Bound registers204 and any methods specified here may perform certain tasks or checks, such as lower/upper bounds checking, when accessing memory using a pointer. These methods may also perform additional checks, such checks of permissions and/or data types. In one embodiment, boundregisters204 are used to specify lower/upper bound values for an address to be used to access memory, enforce the specified bound values, and handle any bound violations.
• In one embodiment, boundregisters204 and methods are used to specify bound values (also referred to as access control information) for pointers, manipulate pointers and/or bound values, as necessary, and enforce bound values as part of memory access. A novel fine grain access control or bounds check technique usingbounds registers204 is employed. The method/apparatus/system employed here directly associates bound values to pointers or pointer values (as opposed to having controls in page table or segment registers) that assert access control for a memory region as opposed to the data objects in the general purpose (e.g., integer) registers202 to facilitate fine grain bound checks of bound values stored in the bound registers204.
• As aforementioned, in one embodiment, one or more existing (but unused) instructions of a program may be used to generate and specify bound values (access control information), and to associate bound values to pointers in the default registers202, and instructions/prefixes are further used to indicate theprocessor200 of the enforcement requirements for specific memory access. For example, the default registers202 may include general purpose (integer) registers to be extended to hold bound values including lower and upper bounds on the address to be accessed using pointers contained in the default registers202. Eachdefault register202 is augmented into two additional registers illustrated as boundregisters204 to hold lower and upper bound values for the pointer in adefault register202.
• When a default (integer) register202 with a valid descriptor is used to access memory, theprocessor200 enforces bounds in the descriptor against the memory address being accessed. Theprocessor200 may then add the operand size to the pointer in thedefault register202 to check against the specified upper bound. For example, if the access was four bytes, theprocessor200 may add 4 to the pointer and compare that number against specified the upper bound. Hence, for bound checking, the operand size is one of implicit operand that is used to check bounds. For example, a bound value may be specified in the base register of a processor-architecture's addressing mode; however, the bound may also be specified in an index register.
• In order to ensure interoperability and compatibility with the existing ABIs the legacy instructions may invalidate the bounds associated with destination registers according to one embodiment. In another embodiment, where interoperability may not be required, those instructions are selected for which bounds are not applicable and the bounds (or metadata) may be invalidated forbounds registers204 associated with destination registers. Also, any instruction that needs to propagate bound values from a specific source register to a specific destination register may add a propagation prefix to the instruction. This way, the embodiments provide a fine grain control over bound values, and also theprocessor200 gets to ensure that any legacy instructions clear descriptors to prevent dangling descriptors. However, in other embodiments, where this interoperability or compatibility may not be necessary, these conditions can be relaxed or ignored.
• In one embodiment, bound checks are applied to any number of memory instructions and may be explicitly specified via a prefix of an instruction, such as a segment override prefix. When an instruction has the segment override prefix which is the same as its default segment, the instruction may enforce the bound check while, in another embodiment, a completely new prefix may be added to the instruction. In one embodiment, an explicit bounds check instruction may be allowed, where such an instruction would have two or three operands. In a two operand form, the first operand is used as a memory address and the second operand is used as an access size (byte, word etc.). Bounds from the base or index register of the memory address are used for the check. In a three operand form, a particular bound register of the boundregisters204 may also be explicitly specified along with the contents of the first two operands. Using the technique described in this figure, operations such as preventing buffer overflow, runtime type checking of code (e.g., C/C++ code, etc.) may be performed.
• FIG. 4 illustrates an embodiment of aprocess400 for managing metadata using bound registers as extension registers of default registers. Atprocessing block402, an instruction of a software program is fetched and decoded in a default register at a processor and if the instruction is a legacy instruction, the legacy instruction is executed atprocessing block404. Atprocessing block406, pointer arithmetic is performed (e.g., adding, subtracting, incrementing, etc.) and the upper and lower bounds of the pointer contained in the default register are preserved in the bounds register that is an extension register of the default register atprocessing block408. If the legacy instruction is not a pointer arithmetic instruction, the bounds are invalidated atprocessing block410.
• If the instruction is CMPKDSC, lower and upper bounds of the address to be accessed using a pointer contained in a default register are computed atprocessing block412 and these bounds are then copied to a bounds register atprocessing block414. If the instruction relates to legacy load, atdecision block416, a determination is made as to whether there has been a bounds violation on legacy load. If yes, an exception is thrown atprocessing block420 and if not, a legacy load operation is performed atprocessing block418. Similarly, if the instruction relates to legacy store, atdecision block422, a determination is made as to whether there has been a bounds violation on legacy store. If yes, an exception is thrown atprocessing block420 and if not, a legacy store operation is performed atprocessing block424.
• If the instruction relates to checked pointer (e.g., a pointer which may have bounds associated with it) load, atdecision block426, a determination is made as to whether there has been a bounds violation on pointer load. If yes, an exception is signaled atprocessing block420 and if not, an address is computed for the descriptor in memory atprocessing block428. Atdecision block430, a determination is made as to whether there are valid bounds of the address in the memory. If yes, the valid bounds from memory for the default register are copied to the bounds register serving as an extension of the default register atprocessing block432. If not, the bounds are regarded as invalid atprocessing block410.
• Similarly, if the instruction being fetched and decoded atprocessing block402 relates to checked pointer store, atdecision block434, a determination is made as to whether there has been a bounds violation on the pointer store. If yes, an exception is signaled atprocessing block420 and if not, an address for the descriptor in memory is computed atprocessing block436. Atdecision block438, a determination is made as to whether there is any memory available to store bounds of the address. If not, an exception is signaled atprocessing block420 and if yes, the bounds are stored from the bound register to the memory descriptor atprocessing block440. The processes (from processingblocks410,414,418,424,432,440) end attransition block442.
• FIG. 5 illustrates an embodiment of aprocess500 for bound check. Atprocessing block502, the lowest address to be accessed is computed using an available instruction of a software program. Atprocessing block504, the highest address to be accessed is computed. Atprocessing block506, a calculation is performed. The calculation determines if the lowest address to be accessed is greater than or equal to the lower bound provided in the bounds register and if the highest address to be accessed is lower than or equal to the upper bound in the bounds register; if so, no violation is detected atprocessing block510. If not, a bounds violation is detected and the process ends atprocessing block508.
• FIG. 6 illustrates an embodiment of aprocess600 for managing metadata using bound registers as extension registers of default registers having data associated with the metadata. Atprocessing block602, a set of bounds register is added to the process that already has a set of default registers. A default register may include a general purpose (integer) register and be used to hold data, pointers, etc. The set of bounds registers may include (hardware-based) registers which are capable of holding metadata (such as bounds) relating to the data present in the default registers. In one embodiment, for each default register two bounds registers are added to the processor such that one of them holds a lower bound and the other holds an upper bound corresponding to the address being used by a pointer in the default register to access memory. Atprocessing block604, computation of the lower and upper bounds associated with an address is performed using computation logic at the processor using one or more instructions of a software program that are available but unused.
• Atdecision block606, using the computation logic, a determination is made as to whether the computed bounds are valid. In one embodiment, the computed bounds are violated if the computed lower bound is lesser than the lower bound already present in the bounds register or if the computed upper bound is greater than the upper bound already present in the bounds registers. This is referred to as widening of the bounds. If the bounds are determined to be widened, an exception is thrown atprocessing block608 and the bounds are not stored in the bounds registers. If, however, the bounds are not widened, they are stored in the lower and upper bounds registers atprocessor block610.
• Embodiments of the present invention include providing new/extension registers of default (general purpose) registers such that the extension registers to hold metadata, using an instruction to create metadata in the extension resisters, load/store metadata from/to memory into the extension registers, checking of memory access against metadata, performing invalidation of metadata by those instructions that do not support metadata for results, and providing implicit association of metadata with default registers. Further, performing instruction encoding using previously implemented NOPs for the new instructions which allows for a software program to work correctly on processors that do not implement this invention correctly without the benefit of metadata; while; on the processors that implement the new technology and use metadata, the same software delivers benefits.
• Moreover, embodiments of the present invention define metadata that specifies bounds for a pointer in the general purpose register. When a pointer is used to reference memory, the processor computes the range of addresses that are to be accessed by an instruction and compares those addresses against the bounds in extension (bounds) registers to ensure that the memory locations accessed by any of the instructions are entirely within those bounds. Any attempt to access a memory location outside the bounds is prevented and in such a case, the processor signals exception to allow the software program to handle this condition. An embodiment includes an address translation function to translate the address where a pointer is stored to the address where the metadata is to be stored in the memory such that the metadata associated with a memory address can be loaded or stored into the extension registers. This address translation technique allows the metadata to be stored at a location (e.g., new/extension/bounds register) that is apart from the location where the corresponding data is stored (e.g., default/general purpose register). For brevity, clarity and consistency, “bounds” are used here as an example of what metadata might include (hence, the term “bounds registers”), but it is contemplated that metadata may include any number of corresponding content associated with data. For example, metadata may include any number of data type [<datatype>] (e.g., integer, floating point, etc.) at extension registers relevant to and associated with the corresponding data at default registers. In another embodiment, it is pointer to type table for complex datatypes, such as structures. In other embodiment, the metadata might include memory access rights, such as read only, read/write, etc.
• FIG. 7 illustrates an embodiment of a computing system (e.g., a computer). The exemplary computing system ofFIG. 7 includes: 1) one ormore processor701 at least one of which may include features described above; 2) a memory control hub (MCH)702; 3) a system memory703 (of which different types exist such as double data rate RAM (DDR RAM), extended data output RAM (EDO RAM) etc.); 4) acache704; 5) an input/output (I/O) control hub (ICH)705; 6) agraphics processor706; 7) a display/screen707 (of which different types exist such as Cathode Ray Tube (CRT), Thin Film Transistor (TFT), Liquid Crystal Display (LCD), DPL, etc.; and 8) one or more I/O devices708.
• The one ormore processors701 execute instructions in order to perform whatever software routines the computing system implements. The instructions frequently involve some sort of operation performed upon data. Both data and instructions are stored insystem memory703 andcache704.Cache704 is typically designed to have shorter latency times thansystem memory703. For example,cache704 might be integrated onto the same silicon chip(s) as the processor(s) and/or constructed with faster static RAM (SRAM) cells whilstsystem memory703 might be constructed with slower dynamic RAM (DRAM) cells. By tending to store more frequently used instructions and data in thecache704 as opposed to thesystem memory703, the overall performance efficiency of the computing system improves.
• System memory703 is deliberately made available to other components within the computing system. For example, the data received from various interfaces to the computing system (e.g., keyboard and mouse, printer port, Local Area Network (LAN) port, modem port, etc.) or retrieved from an internal storage element of the computer system (e.g., hard disk drive) are often temporarily queued intosystem memory703 prior to their being operated upon by the one or more processor(s)701 in the implementation of a software program. Similarly, data that a software program determines should be sent from the computing system to an outside entity through one of the computing system interfaces, or stored into an internal storage element, is often temporarily queued insystem memory703 prior to its being transmitted or stored.
• TheICH705 is responsible for ensuring that such data is properly passed between thesystem memory703 and its appropriate corresponding computing system interface (and internal storage device if the computing system is so designed). TheMCH702 is responsible for managing the various contending requests forsystem memory703 access amongst the processor(s)701, interfaces and internal storage elements that may proximately arise in time with respect to one another.
• One or more I/O devices708 are also implemented in a typical computing system. I/O devices generally are responsible for transferring data to and/or from the computing system (e.g., a networking adapter); or, for large scale non-volatile storage within the computing system (e.g., hard disk drive).ICH705 has bi-directional point-to-point links between itself and the observed I/O devices708.
• Portions of various embodiments of the present invention may be provided as a computer program product, which may include a computer-readable medium having stored thereon computer program instructions, which may be used to program a computer (or other electronic devices) to perform a process according to the embodiments of the present invention. The machine-readable medium may include, but is not limited to, floppy diskettes, optical disks, compact disk read-only memory (CD-ROM), and magneto-optical disks, ROM, RAM, erasable programmable read-only memory (EPROM), electrically EPROM (EEPROM), magnet or optical cards, flash memory, or other type of media/machine-readable medium suitable for storing electronic instructions.
• In the foregoing specification, the invention has been described with reference to specific exemplary embodiments thereof. It will, however, be evident that various modifications and changes may be made thereto without departing from the broader spirit and scope of the invention as set forth in the appended claims. The Specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense.

Claims (20)

1. A method comprising:

expanding default registers of a first processor into metadata registers on the first processor of a computer system, the default regiters having stored thereon data; and

separately storing metadata on the metadata registers, the metadata associated with the data.

2. The method ofclaim 1, wherein the metadata includes bounds associated with addresses being used by pointers representing the data in the default registers to access memory, the method comprising:

computing a lower bound and an upper bound of an address being used by a pointer in a default register to access the memory; and

storing the lower and upper bounds in a metadata register.

3. The method ofclaim 1, further comprising performing automatic bounds checking for memory accesses that utilize one or more default registers that are associated with one or more bounds registers that contain one or more of the bounds.

4. The method ofclaim 1, further comprising employing an existing non-operational (NOP) instruction to generate a set of bounds and perform bounds checking of the set of bounds to preserve inter-operability and compatibility of the metadata registers if they reside on a second processor that does not operate with the metadata, and such that an enabled software program operates on the first and second processors.

5. The method ofclaim 1, further comprising associating a prefix to an instruction being used to create and propagate a set of bounds associated with destination registers and using the prefix with existing memory operations to explicitly perform bounds checking of the set of bounds, the prefix including a NOP prefix or a new prefix, wherein the NOP prefix is employed to facilitate an enabled software program to operate on the first and second processors.

6. The method ofclaim 1, further comprising executing an instruction including a legacy instruction to perform pointer arithmetic to propagate a set of bounds associated with a source pointer to a destination pointer, wherein a default register including a general purpose register having the source and destination pointers, wherein the set of bounds are invalidated if the pointer arithmetic is not performed.

7. The method ofclaim 6, further comprising adding the new prefix to the set of bounds being propagated from the source pointer to the destination pointer, wherein the new prefix indicates the propagation of the set of bounds from the source pointer to the destination pointer.

8. The method ofclaim 6, further comprising performing legacy load operation or legacy store operation if a bounds violation relating to the set of bounds is detected.

9. The method ofclaim 1, further comprising computing an addresses range associated with memory accesses, and comparing the address range having a lowest address and a highest address relating to a pointer with the lower and upper bounds already in the metadata registers to prevent any memory accesses that are outside a bound range of the lower and upper bounds.

10. The method ofclaim 9, wherein preventing comprises avoiding bounds violation, the bounds violation occurs if the lowest address is lesser than the lower bound already in the metadata registers or the highest address is greater than the upper bound already in the metadata registers.

11. The method ofclaim 1, wherein expanding comprises generating at least two metadata registers corresponding to each of the default registers and placing the metadata registers on the processor, wherein a first of the at least two metadata registers to store the lower bound and a second of the at least two metadata registers to store the upper bound.

12. A processor comprising:

default registers being expanded into metadata registers on the processor, the default registers to store data; and

the metadata registers to store metadata associated with the data, the metadata being stored at the metadata registers separately from the data being stored at the default registers.

13. The processor ofclaim 12, wherein the metadata includes bounds associated with addresses being used by pointers representing the data in the default registers to access memory, the processor comprising:

logic to compute a lower bound and an upper bound of an address being used by a pointer in a default register to access the memory; and

a metadata register to store the lower and upper bounds.

14. The processor ofclaim 12, wherein at least two metadata registers are generated corresponding to each of the default registers, wherein a first of the at least two metadata registers to store the lower bound and a second of the at least two metadata registers to store the upper bound.

15. A computer system comprising:

a dynamic random access memory chip; and

a processor having

default registers being expanded into metadata registers on the processor, the default registers to store data, and

the metadata registers to store metadata associated with the data, the metadata being stored at the metadata registers separately from the data being stored at the default registers.

16. The computer system ofclaim 15, wherein the metadata includes bounds associated with addresses being used by pointers representing the data in the default registers to access memory, the processor comprising:

logic to compute a lower bound and an upper bound of an address being used by a pointer in a default register to access the memory; and

a metadata register to store the lower and upper bounds.

17. The computer system ofclaim 15, wherein at least two metadata registers are generated corresponding to each of the default registers, wherein a first of the at least two metadata registers to store the lower bound and a second of the at least two metadata registers to store the upper bound.

18. A machine-readable medium comprising instructions that when executed, cause a machine to:

expand default registers of a processor into metadata registers on the processor of a computer system, the default registers having stored thereon data; and

separately store metadata on the metadata registers, the metadata associated with the data.

19. The machine-readable medium ofclaim 18, wherein the metadata includes bounds associated with addresses being used by pointers representing the data in the default registers to access memory, the instructions that when executed, further cause the machine to:

compute a lower bound and an upper bound of an address being used by a pointer in a default register to access the memory; and

store the lower and upper bounds in a metadata register.

20. The machine-readable medium ofclaim 18, wherein the instructions that when executed, further cause the machine to perform automatic bounds checking for memory accesses that utilize one or more default registers that are associated with one or more bounds registers that contain one or more of the bounds

Images